Chapter 10 Formal Semantics

10.1 Introduction to a very technical and highly formalized

approach

– Formal semantics or logical semantics is the general approach of a family of

theories (denotational) which use logic in semantic analysis.
– Other names focus on particular aspects of this general approach: truth
conditional semantics, model-theoretic semantics, and Montague
Grammar.
– This approach elaborates further the use of truth, truth conditions, and
logic (chapter 4). There we saw the notion of truth, borrowed from logic,
and the formalism of propositional logic to characterize semantic
relations like entailment.
– How further tools from logic can be used to help characterize aspects of
sentence-internal semantics?
– An important philosophical divide in semantics: between
representational and denotational approaches to meaning.
– For semanticists like Jackendoff semantic analysis involves discovering
the conceptual structure which underlies language. For such linguists
the search for meaning is the search for mental representations.
– For formal semanticists on the other hand a primary function of
language is that it allows us to talk about the world around us. When
communicating with others and in our own internal reasoning we use
language to describe, or model, facts and situations. From this
perspective, understanding the meaning of an utterance is being able to
match it with the situation it describes. Hence the search for meaning,
from the denotational perspective, is the search for how the symbols
of language relate to reality.
– How is this relation characterized?
Formal semanticists employ the correspondence theory of truth.
Speakers are (held to be) aware of what situation an utterance, like It´s
raining in London, describes and able to tell whether the utterance and
the situation match up or correspond. A successful match is called true;
an unsuccessful match is false. The listener who understands the
sentence is able to determine the truth conditions of the uttered
sentence, that is, know what conditions in the world would make the
sentence true.
– In the basic version of this approach used in logic there are no almosts or
nearlys: an utterance either describes a situation, and is therefore true of
that situation, or not, in which case it is false.
– The Correspondence Theory (using the notion of truth) says that
a proposition, P, is true if and only if the (possible) state of affairs … is
as P asserts it to be. It defines “truth” as a property which propositions
have just when they “correspond” to the (possible) states of affairs
whose existence they assert.
– Some objections: this seems to apply just to statements, since
intuitively it is hard to see how other utterance types like questions and
orders can be viewed as descriptions of situations. Yet, many utterances
are not statements.
– On a more general level the idea of correct or incorrect matches
seems to remove the subjectivity (degree of certainty) of the speaker,
the speaker attitudes to, or confidence in, a proposition. We describe
such ranges with terms like modality and evidentiality.
– Some advantages:
– Using logical expressions as a semantic metalanguage enables
economy and formality and the benefits of precision; explicit
relations between logical symbols and what they represent and the
effects of combining symbols.
– Denotational approaches, if successful, have another advantage: they
escape the problem of circularity.
– Example of Circularity: to know the meaning of a word = knowing the
definition e.g. coffee = a beverage consisting of an infusion of ground
coffee beans, we need to know the meaning of the words making up the
definition (infusion, coffee beans)! This involves giving further
definitions… Where would this process stop?
– Formal semanticists do translate a natural language like English into
a second, logical language, but this translation is only part of the
semantic analysis. This logical language is then semantically grounded
by tying it to real-world situations. The aim of a denotational approach
is not just to convert between representations: it seeks to connect
language to the world.
– Denotational approaches allow us to see more clearly the connection
between human languages and the simpler signs systems of primates
like monkeys, baboons and chimpanzees. These systems are clearly
referential: primates often have distinct conventional signs for different
types of predators like eagles, snakes, or big cats. Perhaps this basic
matching between a symbol and entities in the environment was the
starting point for human languages.
 10.2 Model-Theoretical Semantics
A modern approach to linguistic description
– An important element in this theory is a model, a formal structure
representing linguistically relevant aspects of a situation.
– In such an approach semantic analysis consists of three stages:
1. a translation from a natural language like English into a logical
language whose syntax and semantics are explicitly defined.
2. the establishment of a mathematical model of the situations that the
language describes.
3. a set of procedures for checking the mapping between the expressions
in the logical language and the modeled situations. Essentially these
algorithms check whether the expressions are true or false of the
modeled situations.
10.3.1 Translating English into a Logical Metalanguage
– The basic idea is that we can translate from a sentence in an individual
language like English into an expression in a universal metalanguage.
One such metalanguage is predicate logic.
– Predicate logic investigates sentence connectives in propositional
logic and goes on to investigate the internal structure of sentences, for
example the truth-conditional effect of certain words like the English
quantifiers all, some, one, and so on. A set of logical connectives parallels
some uses of English expressions like and, or, if … then, and not. For each
connective, its symbol, an example of its syntax, that is, how it combines
with sentence constants p, q:
– Connectives in our translations into predicate logic:
Connective Syntax English
¬ ¬p it is not the case that p
∧ p∧q p and q
∨ p∨q p and/or q
∨e p ∨e q p or q but not both
→ p→q if p, then q
≡p ≡q p if and only if q

10.3.2 Simple statements in predicate logic

– In simple statements like:
a. Moll is sleeping. b. Bill smokes (intransitive sentences)
we can identify a subject-predicate structure where the subject is a
referring expression (Moll, Bill) and the predicate tells us something
about the subject (is asleep, smokes). The predicate logic assigns
different roles to these two elements: the predicate is treated as a
skeletal function which requires the subject argument to be complete.
– Our first step is to represent the predicate by a capital predicate letter,
for example: is asleep: A , smokes: S
The subject argument can be represented by a lower-case letter (usually
chosen from a to t and called an individual constant), for example: Moll:
m Bill: b
– The convention is that predicate logic forms begin with the predicate,
followed by the subject constant. Thus :
a.Moll is asleep: A(m) b. Bill smokes: S(b)
– We can use variables (lower case letters from the end of the alphabet: w,
x, y, z), for example: x is asleep: A(x) , y smokes: S(y)
– The verbs in transitive sentences require more than one nominal.
These predicates are identified as relations between the arguments:
Bill resembles Eddie: R(b, e) Tom adores Libby: A(t, l)
Pete is crazier than Ryan: C(p, r)
– The order of constant terms after the predicate letter is significant: it
mirrors English sentence structure in that the subject comes before the
object.
– Three-place relations; Fatima prefers Bill to Henry: P(f, b, h)
– In our examples so far we have included the English sentence and the
logical translation. Alternatively, we can keep track of what the letters in
the logical form correspond to by providing a key, for example: P(f, b, h)
Key: P: prefer f: Fatima b: Bill h: Henry
– Negative and compound sentences by making use of the connectives :
Mayte doesn't jog: ¬J(m)
Fred smokes and Kate drinks: S(f) ∧ D(k)
If Bill drinks, Jenny gets angry: D(b) → A(j)
– We can represent complex sentences containing relative clauses by
viewing them as a form of conjunction, that is by using ∧ “and,” as in :
Carrick, who is a millionaire, is a socialist: M(c) ∧ S(c)
Emile is a cat that doesn't purr: C(e) ∧ ¬P(e)
Jean admires Robert, who is a gangster: A(j, r) ∧ G(r)
10.3.3 Quantifiers in predicate logic
– Quantification In English for example quantifiers include words like
one, some, a few, many, a lot, most, and all.
– A student wrote a paper. A few students wrote a paper. Many students
wrote a paper. Most students wrote a paper. All students wrote a
paper. Every student wrote a paper. No student wrote a paper.
– Frege's proposal: statements containing quantifiers be divided into two
sections:
the quantifying expression, which gives the range of the
generalization; and the rest of the sentence (the generalization),
which will have a place-holder element, called a variable, for the
quantified nominal.
– Quantifiers all and every are represented in predicate logic by the
universal quantifier, symbolized as ∀
Every student wrote a paper ∀x (S(x) → W(x, p))
For every thing x, if x is a student then x wrote a paper.
The universal quantifier fixes the value of x as every thing; the
expression in parentheses is the generalization. By itself the
generalization is an incomplete proposition, called an open proposition:
until the value of x is set for some individual(s) the expression cannot be
true or false. The quantifier serves to set the value of x and close
the proposition.
– Expressions with the universal quantifier ∀can be paraphrased in
English by all or every as in All students wrote a paper or Every student
wrote a paper.
– The quantifier phrase can be associated with different positions in the
predicate:
– a. Every student knows the professor: ∀x (S(x) →K(x, p))
– b. The professor knows every student: ∀x (S(x) →K(p, x))
– The quantifying expression is said to bind the variable in the predicate
expression; and the predicate expression is said to be the scope of the
quantifier.
– The quantifier some.
– Some is represented in predicate logic by the existential quantifier,
symbolized as ∃: ∃x (S(x) ∧ P(s, e))
There is (at least) one thing x such that x is a student and x wrote a
paper.
– We can paraphrase such expressions in English by using noun phrases
like a student, some student, and at least one student. The existential
quantifier can also be associated with different positions in the
predicate:
(At least) One student kissed Kylie: ∃x (S(x) ∧ K(x, k))
Kylie kissed (at least) one student: ∃x (S(x) ∧ K(k, x))
– Once again the existential quantifier is said to bind the variable and the
predicative expression is described as the scope of the quantifier.
The English determiner no can be represented by a combination of the
existential quantifier and negation, as shown below:
No student wrote a paper ¬∃x (S(x) ∧ W(x, p))
It is not the case that there is a thing x such that x is a student and x
wrote a paper, There is no x such that x is a student and x wrote a
paper.
– Another way of representing this is by using the material implication:
∀x (S(x) → ¬P(x, e))
For every thing x, if x is a student then it is not the case that x wrote a
 paper.
To sum up: the syntax of the predicate logic includes the vocabulary of
symbols and the rules for the formation of logical formulae.
The symbols of predicate logic Predicate letters: A, B, C, etc. Individual
constants: a, b, c, etc.
Individual variables: x, y, z, etc.
Truth functional connectives: ¬, ∧, ∨, , →, ≡ Quantifiers: ∀, ∃
– The rules for creating logical formulae
a. Individual constants and variable are terms.
b. If A is an n-place predicate and t1…tn are n terms, then A(t1…tn) is a
formula.
c. If φ is a formula, then ¬φ is a formula.
d. If φ and ψ are formulae, then (φ ∧ ψ), (φ ∨ ψ), (φ ψ), (φ → ψ), (φ ≡
ψ) are all formulae.
– e. If φ is a formula and x is a variable, then ∀xφ, and ∃xφ are formulae.
10.3.4 Some advantages of predicate logic translation
– The predicate logic is used by logicians to demonstrate the validity of
arguments and reasoning. Thus in addition to a syntax and semantics,
the logical languages requires rules of inference.
– The representation of quantifiers clarifies some ambiguities found in
natural languages. One of these is scope ambiguity, which can occur
when there is more than one quantifier in a sentence. For example the
English sentence Everyone loves someone has two interpretations
paraphrased in Everyone has someone they love ∀x∃y (L(x, y))
The formula: for every person x, there is some person y that they love.
The universal quantifier comes leftmost and therefore contains the
existential quantifier having wide scope.
– There is some person who is loved by everyone.∃y∀x (L(x, y))
We have the reverse: the existential quantifier contains the universal in
its scope and therefore takes wide scope. Thus the scope of one
quantifier may be contained within the scope of another.
– Negative words, like English not, also display scope over predication and
a second advantage is that it allows us to disambiguate some sentences
which contain combinations of quantifiers and negation.
– Everybody didn't visit Paco , for example, can have the two
interpretations:
For every person x, it's not the case that x visited Paco.
∀x ¬(V (x,p))
It's not the case that every person x visited Paco. ¬∀x (V (x, p))
As we can see, the ambiguity is clearly distinguished in the predicate
logic translations.
 10.4 The Semantics of the Logical Metalanguage
10.4.1 Introduction
– The aim of this approach is to devise a denotational semantics.
Translating from an English sentence into a logical formula is not
enough: we then have to relate this second set of symbols to something
outside – the situation described.
– To do this we need to add three further elements:
1. a semantic interpretation for the symbols of the predicate logic;
2. a domain: this is a model of a situation which identifies the
linguistically relevant entities, properties and relations; and
3. a denotation assignment function: this is a procedure, or set of
procedures, which match the logical symbols for nouns, verbs, etc. with
the items in the model that they denote. This function is also sometimes
called a naming function.
– The domain and naming function are together called a model.
10.4.2 The semantic interpretation of predicate logic
symbols
– whole sentences, constant terms, and predicates.
Sentences
– Following the correspondence theory of truth we take the denotatum of
a whole sentence to be the match or lack of match with the situation it
describes. A match will be called true (T), also symbolized by the
numeral 1. A mismatch will be called false (F), symbolized by the
numeral 0. So using a variable v for situations, we might say “a sentence
p is true in situation v,” and symbolize it as [p]v = 1. Here we use square
brackets to symbolize the denotatum of an expression, so [x]v means the
denotatum of x in the situation v. Thus the notation [p]v = 1 means “the
denotatum of p in v is true.” By contrast the expression [p]v = 0 will be
read as “the denotatum of p in v is false” or, equivalently, “the sentence
p is false in situation v.” Since, as we have acknowledged, meaning is
compositional, we want the truth-value of a sentence to be determined
by the semantic value of its parts: the nouns, verbs, connectives, and so
on of which it is constructed.
Individual constant terms
– We will assume the denotation of individual constant terms to be
individuals or sets of individuals in the situation. So if we adopt as our
situation the 1974 world heavyweight title fight between Muhammad Ali
and George Foreman in Zaire, we could use an individual constant term
a to denote Ali, another individual constant f to denote Foreman and a
third, r, to denote the referee in this situation v.

Predicate constants
– We will assume that predicate constants, abbreviated with capital
letters, P, Q, R, and so on, identify sets of individuals for which the
 predicate holds. Thus a one-place predicate like be standing will pick out
the set of individuals who are standing in the situation described. This
can be described in a set theory notation as either {x | … } or {x: … },
both of which can be read as “the set of all x such that …” So a notation
like {x: x is standing in v} can be read as “the set of individuals who are
standing in situation v.”
– Two-place predicates identify a set of ordered pairs: two individuals in a
given order. Thus the predicate punch will pick out an ordered pair
where the first punches the second in v represented in set theory terms
as: {<x, y>: x punches y in v}. Similarly a three-place predicate like hand
to will be represented as
{<x, y, z>: x hands y to z in v}.

10.7 Natural Language Quantifiers and Higher-Order Logic

– One major problem is that there are some common types of quantifiers
which cannot be modeled in this standard form of the predicate
calculus; for example the English quantifier most.
It is impossible to establish most on a par with the universal quantifier
∀ and existential quantifier ∃, using the logical connectives ∧ and →.
a. Most students read a book.
b. Most x(S(x) ∧ R(x, b)) “Most things are students and read books,”
(impossible)
– c. Most x(S(x) → R(x, b)) “Most things are such that if they are
– students they read a book.” (impossible)

10.7.1 Restricted quantifiers

– restricted quantification One step to express the restriction placed on
quantifying determiners by their head nominals : A sentence like All
students are hardworking would be represented in the restricted format by
a. (∀x: S(x)) H(x), compared to the standard format: b. (∀x)(S(x) →
H(x)) Here the information from the rest of the noun phrase is placed
into the quantifying expression as a restriction on the quantifier.
– Similarly One student is hardworking is represented in the restricted
format by a. (∃x: S(x)) H(x), again contrasting with the standard
format: b. (∃x)(S(x) ∧ H(x))
– Restricted quantification helps solve the problem of isomorphism:
it has the advantage that the logical expressions correspond more
closely to natural language expressions. For example, the English noun
phrase all students has a translation into a unitary logical expression: all
students (∀x:S(x)). Most students (Most x: (S(x)); few students as (Few x:
S(x)), and so on.
– In English some quantifiers can stand alone, for example everything,
 everybody, everywhere. These will have to be translated into complex
expressions in predicate logic, as in:
everything every thing (∀x: T(x)) everybody every person (∀x: P(x))
everywhere every location (∀x: L(x))
Everything is either matter or energy: (∀x: T(x)) (M(x) ∨E(x))
Barbara hates everyone: (∀x: P(x)) H(b, x)
Everywhere is dangerous: (∀x: L(x)) D(x)
– As with the universal quantifier, some English words seem to
incorporate an existential quantifier, for example something, someone,
somewhere. These will be expanded in the translation into predicate
logic, as shown below:
something some thing (∃x: T(x)) someone some person (∃x: P(x))
somewhere some location (∃x: L(x))

10.7.2 Generalized quantifiers

– We still need to develop a way to provide a semantic interpretation for
noun phrase formulae like (Most x: (S(x)) most students, (∀x: S(x))
all students, and so on. One proposal is called generalized quantifier
theory.

10.8 Intensionality
10.8.1 Introduction
– Natural languages largely communicate interpretations between
speakers and hearers. For example languages contain a whole range of
verbs which describe different mental states. Instead of a flat statement
S, we can say in English for example:
a. Frank knows that S. b. Frank believes that S.
c. Frank doubts that S. d. Frank regrets that S.
e. Frank suspects that S. f. Frank hopes that S. etc
In these sentences we have a set of propositional attitudes, the choice of
which reflects a difference between certainty and degrees of lack of
certainty.
– In another terminology, sentences which reveal this interpretative or
cognitive behavior are said to be intensional and the property is called
intensionality. These terms are applied whenever linguistic behavior
reveals a relation between an agent and a thought.
– The classical cases are the verbs of propositional attitudes, which in one
terminology are said to form opaque contexts. The term opaque
figuratively describes the fact that the truth or falsity of the subordinate
clause seems to be independent of the truth or falsity of the whole
sentences. We need access to the content of the subject's belief, an extra
level of sense, or in a more recent terminology, intension.
– The challenge for formal semantics is to develop the semantic model to
reflect the interpretation and calculation that is so central to
language.
– Formal devices in certain areas, modality, tense, aspect and verbs of
propositional attitude, where intensionality seems most clearly exhibited
in natural languages
– Epistemic modality, concerns the resources available to the speaker to
express judgment of fact versus possibility. For example,
Allan's scale of modality
a. I know that p. b. I am absolutely certain that p.
c. I am almost certain that p. d. I believe that p.
e. I am pretty certain that p. f. I think that p.
g. I think/believe that p is probable.
h. I think/believe that perhaps p. i. Possibly p.
j. I suppose it is possible that p. k. It is not impossible that p.
l. It is not necessarily impossible that p. m. It is unlikely that p.
n. It is very unlikely that p. o. It is almost impossible that p.
p. It is impossible that p. q. It is not the case that p.
– r. It is absolutely certain that not-p.
– Deontic modality, the second type of modality, allows the expression of
obligation and permission, often in terms of morality and law.
– Deontic modality, has been treated in a similar way: as a projection
from the world as it is to the world as it should be under some moral or
legal code, that is, as the speaker entertaining an idealized world.
Deontic modal operators have been suggested for logic, including
Oφ “obligatorily that φ” “true in all morally or legally ideal worlds”
and Pφ “permitted that φ.” “true in some morally or legally ideal
worlds.”
– All languages allow speakers a range of positions in both of these
aspects. Some of these choices of degree of commitment to the truth of
p derive from the meaning of verbs like believe, know, and so on; others
from negation or from adjectives and adverbs like possible and possibly.
The use of different intonation patterns can add further distinctions.
– Modal logics were developed in response to these facts.
– The simplest approach employs a two-fold division of epistemic
modality into fact versus possibility, or “situation as is” versus “situation
as may be.” One way of discussing this distinction between the actual
and the nonactual is to talk of possible worlds. This notion has been
important in formal semantics to see truth as being relativized to
possible situations, or possible worlds.
– To reflect this, logicians introduce two logical operators
◊ “it is possible that” and □ “it is necessary that.” These can be put in
front of any formula of the predicate logic, that is:
◊φ = it is possible that φ □φ = it is necessary that φ
– The semantic definition:
– □ means “true in all possible worlds” (i.e. No alternatives are
envisaged by the speaker) and
◊ means “true in some possible worlds” (i.e. the speaker does
envisage alternative scenarios).
– Relativizing truth to possible worlds. The formal implications of
this is that truth must be relativized not to one situation but to one
among a series of possible situations (worlds), including the actual
situation (world). This means that our model must be expanded to
include this multiplicity of situations, that is now M = {W, U, F} where,
as before, U = the domain of individuals in a situation, F is the
denotation assignment function, and the new element W is a set of
possible worlds.